From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755231Ab1G1Sm7 (ORCPT ); Thu, 28 Jul 2011 14:42:59 -0400 Received: from merlin.infradead.org ([205.233.59.134]:49381 "EHLO merlin.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754190Ab1G1Sm4 (ORCPT ); Thu, 28 Jul 2011 14:42:56 -0400 Subject: Re: [perf] overflow/perf_count_sw_cpu_clock crashes recent kernels From: Peter Zijlstra To: Vince Weaver Cc: linux-kernel@vger.kernel.org, Paul Mackerras , Ingo Molnar , Arnaldo Carvalho de Melo , stable In-Reply-To: References: Content-Type: text/plain; charset="UTF-8" Date: Thu, 28 Jul 2011 20:47:10 +0200 Message-ID: <1311878830.2617.373.camel@laptop> Mime-Version: 1.0 X-Mailer: Evolution 2.30.3 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Wed, 2011-07-27 at 14:51 -0400, Vince Weaver wrote: > [ 392.504845] ------------[ cut here ]------------ > [ 392.504962] WARNING: at kernel/smp.c:320 smp_call_function_single+0x6c/0xf2() > [ 392.505074] Hardware name: Precision M4500 > [ 392.505181] Modules linked in: acpi_cpufreq cpufreq_conservative mperf cpufreq_powersave cpufreq_userspace cpufreq_stats uinput nouveau snd_hda_codec_hdmi ttm drm_kms_helper mxm_wmi snd_hda_codec_idt iwlagn mac80211 snd_hda_intel snd_hda_codec cfg80211 dell_laptop snd_hwdep video processor ehci_hcd dell_wmi sparse_keymap psmouse sdhci_pci rfkill snd_pcm sdhci thermal_sys pcspkr ac battery wmi serio_raw snd_timer snd_page_alloc evdev i2c_i801 dcdbas button > [ 392.509709] Pid: 2310, comm: overflow_allcou Not tainted 3.0.0 #43 > [ 392.509819] Call Trace: > [ 392.509925] [] ? warn_slowpath_common+0x78/0x8c > [ 392.510144] [] ? perf_exclude_event.part.23+0x31/0x31 > [ 392.510257] [] ? smp_call_function_single+0x6c/0xf2 > [ 392.510369] [] ? task_function_call+0x42/0x4c > [ 392.510476] [] ? update_cgrp_time_from_event+0x2c/0x2c > [ 392.510589] [] ? perf_event_disable+0x45/0x8c > [ 392.510700] [] ? __perf_event_overflow+0xf1/0x1a3 > [ 392.510812] [] ? select_task_rq_fair+0x349/0x574 > [ 392.510924] [] ? perf_ctx_adjust_freq+0x42/0xe6 > [ 392.511038] [] ? sched_clock_cpu+0xb/0xc3 > [ 392.511152] [] ? paravirt_read_tsc+0x5/0x8 > [ 392.511262] [] ? native_sched_clock+0x27/0x2f > [ 392.511366] [] ? perf_event_overflow+0x10/0x10 > [ 392.511476] [] ? perf_swevent_hrtimer+0x9f/0xda > [ 392.511599] [] ? run_posix_cpu_timers+0x23/0x346 > [ 392.511721] [] ? rb_insert_color+0xb1/0xd9 > [ 392.511841] [] ? __run_hrtimer+0xac/0x135 > [ 392.511960] [] ? hrtimer_interrupt+0xdb/0x195 > [ 392.512083] [] ? check_for_new_grace_period.isra.32+0x99/0xa4 > [ 392.512220] [] ? __rcu_process_callbacks+0x72/0x2b7 > [ 392.512345] [] ? hpet_interrupt_handler+0x23/0x2b > [ 392.512469] [] ? handle_irq_event_percpu+0x50/0x180 > [ 392.512592] [] ? __do_softirq+0x13e/0x177 > [ 392.512713] [] ? send_sigio+0x95/0xab > [ 392.512832] [] ? handle_irq_event+0x34/0x52 > [ 392.512952] [] ? handle_edge_irq+0x9f/0xc6 > [ 392.513072] [] ? handle_irq+0x1d/0x21 > [ 392.513192] [] ? do_IRQ+0x42/0x98 > [ 392.513314] [] ? common_interrupt+0x13/0x13 > [ 392.513438] > [ 392.513542] ---[ end trace 12f3f913316a2866 ]--- The below patch is for -stable only, upstream has a much larger patch that contains the below hunk. --- Subject: perf: Fix software event overflow Vince found that under certain circumstances software event overflows go wrong and deadlock. Avoid trying to delete a timer from the timer callback. Cc: stable@kernel.org Reported-by: Vince Weaver Signed-off-by: Peter Zijlstra --- kernel/events/core.c | 7 ++----- 1 files changed, 2 insertions(+), 5 deletions(-) diff --git a/kernel/events/core.c b/kernel/events/core.c index 9efe710..32a6151 100644 --- a/kernel/events/core.c +++ b/kernel/events/core.c @@ -5016,11 +5016,8 @@ static int __perf_event_overflow(struct perf_event *event, int nmi, if (events && atomic_dec_and_test(&event->event_limit)) { ret = 1; event->pending_kill = POLL_HUP; - if (nmi) { - event->pending_disable = 1; - irq_work_queue(&event->pending); - } else - perf_event_disable(event); + event->pending_disable = 1; + irq_work_queue(&event->pending); } if (event->overflow_handler)